基于面向对象特征提取的BP神经网络分类——以武陵源地区为例

Classification of BP Neural Network Based on Object-Based Feature Extraction——Taking Wulingyuan Area as an Example

采用面向对象影像分类与BP神经网络分类相结合的方法,对高分辨率无人机影像进行土地利用分类。利用光谱、形状、纹理、对象间关系等影像特征,通过基于面向对象的方法对影像提取特征进行初步分类,再将初步分类结果应用于BP神经网络,结合原影像数据进行进一步分类,提高分类精度、纠正分类错误。结果表明,该方法最终分类结果达到了88.9%的总体分类精度和0.863的Kappa系数,影像分类结果对比传统影像分类方法的总体精度与Kappa系数均有所提高。

Combined object-based image classification and BP neural network classification to classify high-resolution UAV images for land use. By using the image features such as spectrum, shape, texture and object relationship, the features are extracted from the image by object-based method, and the preliminary classification results are applied to BP neural network and further classified according to the original image data to improve classification accuracy, correct classification error. The results show that the classification results of this method are 88. 9% of the overall classification accuracy and the Kappa coefficient is 0.863. The overall classification of the image classification results and the Kappa coefficient of the traditional image classification method are improved.